Anomaly Detection System for Embedded Devices

Abstract

An anomaly detection system may be configured for an embedded device, such as a microcontroller. The anomaly detection system may be configured to receive a dataset from a sensor, such as a camera, a microphone, or an inertial management unit. The anomaly detection system may extract a plurality of features from the dataset. The plurality of features may be configured to train a neural network model, such as a convolutional classifier, to generate one or more classifications. The anomaly detection system may generate an anomaly score based on the plurality of features. The anomaly detection system may trigger an output based on the anomaly score exceeding a range.

What is claimed is: 
     
         1 . A method, comprising:
 receiving, by an embedded device, a dataset from a sensor;   extracting a plurality of features from the dataset, the plurality of features configured to train a neural network model to generate one or more classifications;   generating an anomaly score based on the plurality of features; and   triggering an output based on the anomaly score exceeding a range.   
     
     
         2 . The method of  claim 1 , further comprising:
 collecting a plurality of datasets from the sensor; and   training a neural network, based on the plurality of datasets, to generate the anomaly score.   
     
     
         3 . The method of  claim 1 , wherein extracting the plurality of features comprises:
 utilizing a pretrained network architecture that implements a neural network.   
     
     
         4 . The method of  claim 1 , wherein extracting the plurality of features comprises:
 utilizing a digital signal processing (DSP) algorithm based on at least one of Mel-filterbank energy (MFE), Mel frequency cepstral coefficients (MFCC), or spectrogram.   
     
     
         5 . The method of  claim 1 , further comprising:
 utilizing a Gaussian mixture model (GMM) to determine that the anomaly score exceeds the range.   
     
     
         6 . The method of  claim 1 , further comprising:
 determining cells associated with the dataset; and   generating a plurality of anomaly scores based on the plurality of features, wherein an anomaly score of the plurality of anomaly scores corresponds to a cell of the cells.   
     
     
         7 . The method of  claim 1 , wherein the dataset from the sensor comprises an image from a camera and a pattern indicated by the image causes the anomaly score to exceed the range. 
     
     
         8 . The method of  claim 1 , wherein the dataset from the sensor comprises audio data from a microphone and a sound indicated by the audio data causes the anomaly score to exceed the range. 
     
     
         9 . The method of  claim 1 , wherein the dataset from the sensor comprises motion data from an inertial management unit (IMU) sensor and a vibration indicated by the motion data causes the anomaly score to exceed the range. 
     
     
         10 . A method, comprising:
 configuring an anomaly detection system for an embedded device to perform the steps of:
 receiving, by the embedded device, a dataset from a sensor; 
 extracting a plurality of features from the dataset, the plurality of features configured to train a neural network model to generate one or more classifications; 
 generating an anomaly score based on the plurality of features; and 
 triggering an output based on the anomaly score exceeding a range. 
   
     
     
         11 . The method of  claim 10 , further comprising:
 configuring the anomaly detection system to train a neural network, based on a plurality of datasets collected from the sensor, to generate the anomaly score.   
     
     
         12 . The method of  claim 10 , further comprising:
 configuring the anomaly detection system to utilize a pretrained network architecture that implements a neural network.   
     
     
         13 . The method of  claim 10 , wherein extracting the plurality of features comprises utilizing a DSP algorithm based on at least one of MFE, MFCC, or spectrogram. 
     
     
         14 . The method of  claim 10 , further comprising:
 configuring the anomaly detection system to utilize a statistical distribution to determine that the anomaly score corresponds to anomalous data.   
     
     
         15 . The method of  claim 10 , further comprising:
 configuring the anomaly detection system to determine cells associated with the dataset, the cells arranged in a grid, and generate a plurality of anomaly scores based on the plurality of features, wherein an anomaly score of the plurality of anomaly scores corresponds to a cell in the grid.   
     
     
         16 . An embedded device, comprising:
 a sensor;   a memory; and   a processor configured to execute instructions stored in the memory to:
 receive a dataset from the sensor; 
 extract a plurality of features from the dataset, the plurality of features configured to train a neural network model to generate one or more classifications; 
 generate an anomaly score based on the plurality of features; and 
 trigger an output based on the anomaly score exceeding a range. 
   
     
     
         17 . The embedded device of  claim 16 , wherein the processor is configured to execute instructions stored in the memory to:
 collect a plurality of datasets from the sensor; and   train a neural network, based on the plurality of datasets, to generate the anomaly score.   
     
     
         18 . The embedded device of  claim 16 , wherein the processor is configured to execute instructions stored in the memory to:
 utilize a pretrained network architecture that implements a neural network.   
     
     
         19 . The embedded device of  claim 16 , wherein the processor is configured to execute instructions stored in the memory to:
 utilize at least one of spectral features or wavelets to extract the plurality of features.   
     
     
         20 . The embedded device of  claim 16 , wherein the processor is configured to execute instructions stored in the memory to:
 utilize a mathematical model to determine that the anomaly score exceeds the range.